Merge branch 'tor-gitlab/mr/284' into ticket2667_043_01

8 files changed, 274 insertions, 31 deletions

diff --git a/changes/ticket2667 b/changes/ticket2667
new file mode 100644
index 0000000000..cc42286ef9
--- /dev/null
+++ b/changes/ticket2667
@@ -0,0 +1,4 @@
+  o Major feature (exit):
+    - Re-entry into the network is now denied at the Exit level to all relays'
+      ORPort and authorities' ORPort+DirPort. This is to help mitigate a series
+      of attacks. See ticket for more information. Closes ticket 2667.
diff --git a/src/core/or/address_set.c b/src/core/or/address_set.c
index d618ffc3d6..fcddc55e9f 100644
--- a/src/core/or/address_set.c
+++ b/src/core/or/address_set.c
@@ -69,3 +69,77 @@ address_set_probably_contains(const address_set_t *set,
 {
   return bloomfilt_probably_contains(set, addr);
 }
+
+/* Length of the item is an address (IPv4 or IPv6) and a 2 byte port. We use
+ * 16 bytes for the address here (IPv6) since we do not know which family
+ * the given address in the item thus in the case of IPv4, the extra bytes
+ * are simply zeroes to accomodate. */
+#define BLOOMFILT_ADDR_PORT_ITEM_LEN (16 + sizeof(uint16_t))
+
+/** Build an item for the bloomfilter consisting of an address and port pair.
+ *
+ * If the given address is _not_ AF_INET or AF_INET6, then the item is an
+ * array of 0s.
+ *
+ * Return a pointer to a static buffer containing the item. Next call to this
+ * function invalidates its previous content. */
+static const uint8_t *
+build_addr_port_item(const tor_addr_t *addr, const uint16_t port)
+{
+  static uint8_t data[BLOOMFILT_ADDR_PORT_ITEM_LEN];
+
+  memset(data, 0, sizeof(data));
+  switch (tor_addr_family(addr)) {
+  case AF_INET:
+    memcpy(data, &addr->addr.in_addr.s_addr, 4);
+    break;
+  case AF_INET6:
+    memcpy(data, &addr->addr.in6_addr.s6_addr, 16);
+    break;
+  case AF_UNSPEC:
+    /* Leave the 0. */
+    break;
+  default:
+    /* LCOV_EXCL_START */
+    tor_fragile_assert();
+    /* LCOV_EXCL_STOP */
+  }
+
+  memcpy(data + 16, &port, sizeof(port));
+  return data;
+}
+
+/** Return a hash value for the given item that the bloomfilter will use. */
+static uint64_t
+bloomfilt_addr_port_hash(const struct sipkey *key,
+                         const void *item)
+{
+  return siphash24(item, BLOOMFILT_ADDR_PORT_ITEM_LEN, key);
+}
+
+/** Allocate and return an addr_port_set_t, suitable for holding up to
+ * max_address_guess distinct values. */
+addr_port_set_t *
+addr_port_set_new(int max_addresses_guess)
+{
+  uint8_t k[BLOOMFILT_KEY_LEN];
+  crypto_rand((void*)k, sizeof(k));
+  return bloomfilt_new(max_addresses_guess, bloomfilt_addr_port_hash, k);
+}
+
+/** Add an address and port pair to the given set. */
+void
+addr_port_set_add(addr_port_set_t *set, const tor_addr_t *addr, uint16_t port)
+{
+  bloomfilt_add(set, build_addr_port_item(addr, port));
+}
+
+/** Return true if the given address and port pair are in the set. Of course,
+ * this is a bloomfilter and thus in rare occasion, a false positive happens
+ * thus the "probably". */
+bool
+addr_port_set_probably_contains(const addr_port_set_t *set,
+                                const tor_addr_t *addr, uint16_t port)
+{
+  return !!bloomfilt_probably_contains(set, build_addr_port_item(addr, port));
+}
diff --git a/src/core/or/address_set.h b/src/core/or/address_set.h
index 33887a7c9e..2c78ab0576 100644
--- a/src/core/or/address_set.h
+++ b/src/core/or/address_set.h
@@ -13,13 +13,14 @@
 #include "lib/cc/torint.h"
 #include "lib/container/bloomfilt.h"
 
+struct tor_addr_t;
+
 /**
  * An address_set_t represents a set of tor_addr_t values. The implementation
  * is probabilistic: false negatives cannot occur but false positives are
  * possible.
  */
 typedef struct bloomfilt_t address_set_t;
-struct tor_addr_t;
 
 address_set_t *address_set_new(int max_addresses_guess);
 #define address_set_free(set) bloomfilt_free(set)
@@ -28,4 +29,19 @@ void address_set_add_ipv4h(address_set_t *set, uint32_t addr);
 int address_set_probably_contains(const address_set_t *set,
                                   const struct tor_addr_t *addr);
 
+/**
+ * An addr_port_set_t represents a set of tor_addr_t values with a uint16_t
+ * port value. The implementation is probabilistic: false negatives cannot
+ * occur but false positives are possible.
+ */
+typedef struct bloomfilt_t addr_port_set_t;
+
+addr_port_set_t *addr_port_set_new(int max_addresses_guess);
+#define addr_port_set_free(s) bloomfilt_free(s)
+void addr_port_set_add(addr_port_set_t *set,
+                       const struct tor_addr_t *addr, uint16_t port);
+bool addr_port_set_probably_contains(const addr_port_set_t *set,
+                                     const struct tor_addr_t *addr,
+                                     uint16_t port);
+
 #endif /* !defined(TOR_ADDRESS_SET_H) */
diff --git a/src/core/or/connection_edge.c b/src/core/or/connection_edge.c
index 37e83ba71e..80b02ccad3 100644
--- a/src/core/or/connection_edge.c
+++ b/src/core/or/connection_edge.c
@@ -4181,6 +4181,31 @@ connection_exit_connect(edge_connection_t *edge_conn)
     return;
   }
 
+  /* Next, check for attempts to connect back into the Tor network. We don't
+   * want to allow these for the same reason we don't want to allow
+   * infinite-length circuits (see "A Practical Congestion Attack on Tor Using
+   * Long Paths", Usenix Security 2009). See also ticket 2667.
+   *
+   * The TORPROTOCOL reason is used instead of EXITPOLICY so client do NOT
+   * attempt to retry connecting onto another circuit that will also fail
+   * bringing considerable more load on the network if so.
+   *
+   * Since the address+port set here is a bloomfilter, in very rare cases, the
+   * check will create a false positive meaning that the destination could
+   * actually be legit and thus being denied exit. However, sending back a
+   * reason that makes the client retry results in much worst consequences in
+   * case of an attack so this is a small price to pay. */
+  if (!connection_edge_is_rendezvous_stream(edge_conn) &&
+      nodelist_reentry_probably_contains(&conn->addr, conn->port)) {
+    log_info(LD_EXIT, "%s:%d tried to connect back to a known relay address. "
+                      "Closing.", escaped_safe_str_client(conn->address),
+             conn->port);
+    connection_edge_end(edge_conn, END_STREAM_REASON_TORPROTOCOL);
+    circuit_detach_stream(circuit_get_by_edge_conn(edge_conn), edge_conn);
+    connection_free(conn);
+    return;
+  }
+
 #ifdef HAVE_SYS_UN_H
   if (conn->socket_family != AF_UNIX) {
 #else
diff --git a/src/feature/nodelist/dirlist.c b/src/feature/nodelist/dirlist.c
index ad3af0a143..883564e080 100644
--- a/src/feature/nodelist/dirlist.c
+++ b/src/feature/nodelist/dirlist.c
@@ -54,13 +54,18 @@ static smartlist_t *fallback_dir_servers = NULL;
 static void
 add_trusted_dir_to_nodelist_addr_set(const dir_server_t *dir)
 {
+  tor_addr_t tmp_addr;
+
   tor_assert(dir);
   tor_assert(dir->is_authority);
 
-  /* Add IPv4 and then IPv6 if applicable. */
-  nodelist_add_addr4_to_address_set(dir->addr);
+  /* Add IPv4 and then IPv6 if applicable. For authorities, we add the ORPort
+   * and DirPort so re-entry into the network back to them is not possible. */
+  tor_addr_from_ipv4h(&tmp_addr, dir->addr);
+  nodelist_add_addr_to_address_set(&tmp_addr, dir->or_port, dir->dir_port);
   if (!tor_addr_is_null(&dir->ipv6_addr)) {
-    nodelist_add_addr6_to_address_set(&dir->ipv6_addr);
+    /* IPv6 DirPort is not a thing yet for authorities. */
+    nodelist_add_addr_to_address_set(&dir->ipv6_addr, dir->ipv6_orport, 0);
   }
 }
 
diff --git a/src/feature/nodelist/nodelist.c b/src/feature/nodelist/nodelist.c
index 6c7fd6d0f9..5df8e63073 100644
--- a/src/feature/nodelist/nodelist.c
+++ b/src/feature/nodelist/nodelist.c
@@ -135,6 +135,10 @@ typedef struct nodelist_t {
   /* Set of addresses that belong to nodes we believe in. */
   address_set_t *node_addrs;
 
+  /* Set of addresses + port that belong to nodes we know and that we don't
+   * allow network re-entry towards them. */
+  addr_port_set_t *reentry_set;
+
   /* The valid-after time of the last live consensus that initialized the
    * nodelist.  We use this to detect outdated nodelists that need to be
    * rebuilt using a newer consensus. */
@@ -447,49 +451,62 @@ node_addrs_changed(node_t *node)
 static void
 node_add_to_address_set(const node_t *node)
 {
-  if (!the_nodelist || !the_nodelist->node_addrs)
+  tor_addr_t tmp_addr;
+
+  if (!the_nodelist ||
+      !the_nodelist->node_addrs || !the_nodelist->reentry_set)
     return;
 
-  /* These various address sources can be redundant, but it's likely faster
-   * to add them all than to compare them all for equality. */
+  /* These various address sources can be redundant, but it's likely faster to
+   * add them all than to compare them all for equality.
+   *
+   * For relays, we only add the ORPort in the addr+port set since we want to
+   * allow re-entry into the network to the DirPort so the self reachability
+   * test succeeds and thus the 0 value for the DirPort. */
 
   if (node->rs) {
-    if (node->rs->addr)
-      nodelist_add_addr4_to_address_set(node->rs->addr);
+    if (node->rs->addr) {
+      tor_addr_from_ipv4h(&tmp_addr, node->rs->addr);
+      nodelist_add_addr_to_address_set(&tmp_addr, node->rs->or_port, 0);
+    }
     if (!tor_addr_is_null(&node->rs->ipv6_addr))
-      nodelist_add_addr6_to_address_set(&node->rs->ipv6_addr);
+      nodelist_add_addr_to_address_set(&node->rs->ipv6_addr,
+                                       node->rs->ipv6_orport, 0);
   }
   if (node->ri) {
-    if (node->ri->addr)
-      nodelist_add_addr4_to_address_set(node->ri->addr);
+    if (node->ri->addr) {
+      tor_addr_from_ipv4h(&tmp_addr, node->ri->addr);
+      nodelist_add_addr_to_address_set(&tmp_addr, node->ri->or_port, 0);
+    }
     if (!tor_addr_is_null(&node->ri->ipv6_addr))
-      nodelist_add_addr6_to_address_set(&node->ri->ipv6_addr);
+      nodelist_add_addr_to_address_set(&node->ri->ipv6_addr,
+                                       node->ri->ipv6_orport, 0);
   }
   if (node->md) {
     if (!tor_addr_is_null(&node->md->ipv6_addr))
-      nodelist_add_addr6_to_address_set(&node->md->ipv6_addr);
+      nodelist_add_addr_to_address_set(&node->md->ipv6_addr,
+                                       node->md->ipv6_orport, 0);
   }
 }
 
-/** Add the given v4 address into the nodelist address set. */
+/** Add the given address into the nodelist address set. */
 void
-nodelist_add_addr4_to_address_set(const uint32_t addr)
+nodelist_add_addr_to_address_set(const tor_addr_t *addr,
+                                 uint16_t or_port, uint16_t dir_port)
 {
-  if (!the_nodelist || !the_nodelist->node_addrs || addr == 0) {
-    return;
-  }
-  address_set_add_ipv4h(the_nodelist->node_addrs, addr);
-}
-
-/** Add the given v6 address into the nodelist address set. */
-void
-nodelist_add_addr6_to_address_set(const tor_addr_t *addr)
-{
-  if (BUG(!addr) || tor_addr_is_null(addr) || tor_addr_is_v4(addr) ||
-      !the_nodelist || !the_nodelist->node_addrs) {
+  if (BUG(!addr) || tor_addr_is_null(addr) ||
+      (!tor_addr_is_v4(addr) && tor_addr_family(addr) != AF_INET6) ||
+      !the_nodelist || !the_nodelist->node_addrs ||
+      !the_nodelist->reentry_set) {
     return;
   }
   address_set_add(the_nodelist->node_addrs, addr);
+  if (or_port != 0) {
+    addr_port_set_add(the_nodelist->reentry_set, addr, or_port);
+  }
+  if (dir_port != 0) {
+    addr_port_set_add(the_nodelist->reentry_set, addr, dir_port);
+  }
 }
 
 /** Return true if <b>addr</b> is the address of some node in the nodelist.
@@ -506,6 +523,21 @@ nodelist_probably_contains_address(const tor_addr_t *addr)
   return address_set_probably_contains(the_nodelist->node_addrs, addr);
 }
 
+/** Return true if <b>addr</b> is the address of some node in the nodelist and
+ * corresponds also to the given port. If not, probably return false. */
+bool
+nodelist_reentry_probably_contains(const tor_addr_t *addr, uint16_t port)
+{
+  if (BUG(!addr) || BUG(!port))
+    return false;
+
+  if (!the_nodelist || !the_nodelist->reentry_set)
+    return false;
+
+  return addr_port_set_probably_contains(the_nodelist->reentry_set,
+                                         addr, port);
+}
+
 /** Add <b>ri</b> to an appropriate node in the nodelist.  If we replace an
  * old routerinfo, and <b>ri_old_out</b> is not NULL, set *<b>ri_old_out</b>
  * to the previous routerinfo.
@@ -637,10 +669,13 @@ nodelist_set_consensus(networkstatus_t *ns)
    * v6). Then we add the number of configured trusted authorities we have. */
   int estimated_addresses = smartlist_len(ns->routerstatus_list) *
                             get_estimated_address_per_node();
-  estimated_addresses += (get_n_authorities(V3_DIRINFO & BRIDGE_DIRINFO) *
+  estimated_addresses += (get_n_authorities(V3_DIRINFO | BRIDGE_DIRINFO) *
                           get_estimated_address_per_node());
   address_set_free(the_nodelist->node_addrs);
+  addr_port_set_free(the_nodelist->reentry_set);
   the_nodelist->node_addrs = address_set_new(estimated_addresses);
+  /* Times two here is for both the ORPort and DirPort. */
+  the_nodelist->reentry_set = addr_port_set_new(estimated_addresses * 2);
 
   SMARTLIST_FOREACH_BEGIN(ns->routerstatus_list, routerstatus_t *, rs) {
     node_t *node = node_get_or_create(rs->identity_digest);
@@ -867,6 +902,8 @@ nodelist_free_all(void)
 
   address_set_free(the_nodelist->node_addrs);
   the_nodelist->node_addrs = NULL;
+  addr_port_set_free(the_nodelist->reentry_set);
+  the_nodelist->reentry_set = NULL;
 
   tor_free(the_nodelist);
 }
diff --git a/src/feature/nodelist/nodelist.h b/src/feature/nodelist/nodelist.h
index 87020b81eb..1f111a0b99 100644
--- a/src/feature/nodelist/nodelist.h
+++ b/src/feature/nodelist/nodelist.h
@@ -35,8 +35,10 @@ node_t *nodelist_add_microdesc(microdesc_t *md);
 void nodelist_set_consensus(networkstatus_t *ns);
 void nodelist_ensure_freshness(networkstatus_t *ns);
 int nodelist_probably_contains_address(const tor_addr_t *addr);
-void nodelist_add_addr4_to_address_set(const uint32_t addr);
-void nodelist_add_addr6_to_address_set(const tor_addr_t *addr);
+bool nodelist_reentry_probably_contains(const tor_addr_t *addr,
+                                        uint16_t port);
+void nodelist_add_addr_to_address_set(const tor_addr_t *addr,
+                                      uint16_t or_port, uint16_t dir_port);
 
 void nodelist_remove_microdesc(const char *identity_digest, microdesc_t *md);
 void nodelist_remove_routerinfo(routerinfo_t *ri);
diff --git a/src/test/test_address_set.c b/src/test/test_address_set.c
index 829ecd79e8..4c99008ae3 100644
--- a/src/test/test_address_set.c
+++ b/src/test/test_address_set.c
@@ -182,11 +182,91 @@ test_nodelist(void *arg)
   UNMOCK(dirlist_add_trusted_dir_addresses);
 }
 
+/** Test that the no-reentry exit filter works as intended */
+static void
+test_exit_no_reentry(void *arg)
+{
+  routerstatus_t *rs = NULL; microdesc_t *md = NULL; routerinfo_t *ri = NULL;
+  (void) arg;
+
+  MOCK(networkstatus_get_latest_consensus,
+       mock_networkstatus_get_latest_consensus);
+  MOCK(networkstatus_get_latest_consensus_by_flavor,
+       mock_networkstatus_get_latest_consensus_by_flavor);
+  MOCK(get_estimated_address_per_node,
+       mock_get_estimated_address_per_node);
+  MOCK(dirlist_add_trusted_dir_addresses,
+       mock_dirlist_add_trusted_dir_addresses);
+
+  dummy_ns = tor_malloc_zero(sizeof(*dummy_ns));
+  dummy_ns->flavor = FLAV_MICRODESC;
+  dummy_ns->routerstatus_list = smartlist_new();
+
+  tor_addr_t addr_v4, addr_v6, dummy_addr;
+  tor_addr_parse(&addr_v4, "42.42.42.42");
+  tor_addr_parse(&addr_v6, "1:2:3:4::");
+  memset(&dummy_addr, 'A', sizeof(dummy_addr));
+
+  /* This will make the nodelist bloom filter very large
+   * (the_nodelist->node_addrs) so we will fail the contain test rarely. */
+  addr_per_node = 1024;
+
+  /* After this point the nodelist is populated with the directory authorities
+   * address and ports */
+  nodelist_set_consensus(dummy_ns);
+
+  /* The address set is empty. Try it anyway */
+  tt_assert(!nodelist_reentry_probably_contains(&addr_v4, 244));
+  tt_assert(!nodelist_reentry_probably_contains(&addr_v6, 244));
+
+  /* Now let's populate the network */
+  md = tor_malloc_zero(sizeof(*md));
+  ri = tor_malloc_zero(sizeof(*ri));
+  rs = tor_malloc_zero(sizeof(*rs));
+  crypto_rand(rs->identity_digest, sizeof(rs->identity_digest));
+  crypto_rand(md->digest, sizeof(md->digest));
+  memcpy(rs->descriptor_digest, md->digest, DIGEST256_LEN);
+
+  /* Setup the rs, ri and md addresses. */
+  rs->addr = tor_addr_to_ipv4h(&addr_v4);
+  rs->or_port = 444;
+  tor_addr_parse(&rs->ipv6_addr, "1:2:3:4::");
+  rs->ipv6_orport = 666;
+  ri->addr = tor_addr_to_ipv4h(&addr_v4);
+  tor_addr_parse(&ri->ipv6_addr, "1:2:3:4::");
+  tor_addr_parse(&md->ipv6_addr, "1:2:3:4::");
+
+  /* Add the rs to the consensus becoming a node_t. */
+  smartlist_add(dummy_ns->routerstatus_list, rs);
+  nodelist_set_consensus(dummy_ns);
+
+  /* Now that the nodelist is populated let's do some retry attempts */
+
+  /* First let's try an address that is on the no-reentry list, but with a
+     different port */
+  tt_assert(!nodelist_reentry_probably_contains(&addr_v4, 666));
+  tt_assert(!nodelist_reentry_probably_contains(&addr_v6, 444));
+
+  /* OK now let's try with the right address and right port */
+  tt_assert(nodelist_reentry_probably_contains(&addr_v4, 444));
+  tt_assert(nodelist_reentry_probably_contains(&addr_v6, 666));
+
+ done:
+  routerstatus_free(rs); routerinfo_free(ri); microdesc_free(md);
+  smartlist_clear(dummy_ns->routerstatus_list);
+  networkstatus_vote_free(dummy_ns);
+  UNMOCK(networkstatus_get_latest_consensus);
+  UNMOCK(networkstatus_get_latest_consensus_by_flavor);
+  UNMOCK(get_estimated_address_per_node);
+  UNMOCK(dirlist_add_trusted_dir_addresses);
+}
+
 struct testcase_t address_set_tests[] = {
   { "contains", test_contains, TT_FORK,
     NULL, NULL },
   { "nodelist", test_nodelist, TT_FORK,
     NULL, NULL },
+  { "exit_no_reentry", test_exit_no_reentry, TT_FORK, NULL, NULL },
 
   END_OF_TESTCASES
 };